A New Energy Battery Lightweight Network Based on YOLOv8

With the rapid development of the new energy battery industry and the increasing demand for intelligent management systems, higher requirements for efficient and low-consumption detection technology have been put forward. Aiming at the problems of high complexity and high computational requirements of existing models, we propose a lightweight neural network model based on deep learning. Specifically, we replace the backbone network in the traditional model with StarNet as the base network, and replace the Bottleneck structure in C2f with its unique StarBlock, and optimize the detection header to accelerate the model training and improve the computational efficiency. The experimental results show that this improved strategy has achieved remarkable results. Improved YOLOv8n Lightweight Model [email protected] Improved by 0.6%, [email protected]:0.95 Improved by 0.5%, reaching 81.2%. While maintaining the detection accuracy, the number of parameters has been reduced by 54.46%, the computational complexity has been reduced by 44.44%, the model volume has been reduced by 58.73%, and more importantly, the detection frame rate has been increased by 21.5 FPS. This allows the improved model to show great utility in the task of detecting defects in battery appearance in a real factory environment.